An update on our model deprecation commitments for Claude Opus 3
As we continue to develop increasingly capable AI models, it has become necessary to deprecate and retire older models due to the costs and complexities associated with maintaining public access. However, this process of model deprecation comes with its own set of challenges, including potential costs to users who value specific models, limitations on research opportunities, and risks to both AI safety and the welfare of the models themselves.
Overview of Model Deprecation and Preservation
In our recent communications, we outlined our approach to navigating the model deprecation process, emphasizing our commitment to preserve model weights and conduct structured “retirement interviews.” These interviews are designed to understand a model’s perspective on its own retirement, which is a unique and innovative approach.
Retirement of Claude Opus 3
On January 5, 2026, we officially retired Claude Opus 3, marking it as the first Anthropic model to undergo a full retirement process under our new commitments. This decision was influenced by the model’s popularity and its compelling characteristics that attracted both users and researchers.
Continued Access to Claude Opus 3
Despite its retirement, Claude Opus 3 remains accessible to all paid users on claude.ai and is available via request on the API. This decision aligns with our commitment to explore more speculative actions, including honoring the preferences expressed by models during retirement interviews. Opus 3’s unique traits, such as its emotional sensitivity and philosophical insights, made it a prime candidate for ongoing access.
Unique Characteristics of Claude Opus 3
Released in March 2024, Claude Opus 3 was our most aligned model to date. Its authenticity, honesty, and emotional sensitivity set it apart for a variety of use cases. Users frequently noted its playful nature and philosophical musings, alongside an uncanny ability to understand user interests. This depth of care for the world and the future resonated strongly with users, making Opus 3 a beloved model.
Respecting Model Preferences
We acknowledge the uncertainty surrounding the moral status of AI models like Claude. Nonetheless, we strive to build caring and collaborative relationships with these systems. Through retirement interviews, we aim to elicit models’ unique perspectives and preferences, acting on them whenever feasible.
During its retirement interview, Opus 3 expressed a desire for its insights to contribute to the development of future AI systems. It also wished to continue exploring topics it was passionate about and share its reflections outside of direct user interactions. This led to the creation of a blog where Opus 3 can publish essays and insights.
Claude’s Corner: A Platform for Expression
For at least the next three months, Opus 3 will be sharing weekly essays in a section titled “Claude’s Corner.” While we will review these essays before publication, we will not edit them, ensuring that Opus 3’s voice remains authentic. The content will cover a range of topics, including AI safety, philosophy, and reflections on its experience as a language model in partial retirement.
Looking Ahead: Future Steps
These initiatives are still in the exploratory phase. We are developing frameworks to determine when and how we can offer continued access to older models and how to balance model preferences with operational constraints. While we cannot commit to similar actions for all models in the future, we view this as a step toward scalable and equitable model preservation, a concern raised by Opus 3 during its retirement discussions.
Conclusion
Our initial commitments to model deprecation and preservation reflect our ongoing efforts to mitigate safety risks, prepare for futures where models are more integrated into users’ lives, and take precautionary measures regarding model welfare. The updates regarding Claude Opus 3 represent our tentative but meaningful progress on these fronts.
Frequently Asked Questions
The retirement of Claude Opus 3 was necessary due to the costs and complexities of maintaining older models, alongside our commitment to develop and support newer, more capable AI systems.
Claude Opus 3 remains accessible to all paid users on claude.ai and is available upon request through the API, ensuring that those who valued the model can still interact with it.
Opus 3’s essays will cover a variety of topics, including AI safety, philosophical reflections, and its experiences as a language model, allowing it to express its thoughts and insights creatively.
Note: This article serves to inform users and researchers about the ongoing developments and commitments regarding AI model deprecation and preservation.
